Who is responsible for an AI face? Three principles and what they mean in practice

Generative AI creates realistic people in minutes. The ethical debate that follows is older than the technology, and it is gaining weight: human dignity, deepfakes, bias in training data, the use of faces and voices that nobody asked for. Much of this is argued in principle. For a company that wants to publish an AI campaign tomorrow, the question is more practical: which of these principles can I actually act on today?

Three principles appear in almost every guideline, from the EU AI Act to corporate codes of conduct: transparency, self-determination and responsibility. All three turn out to be very concrete the moment a real person is involved.

Transparency: say that it is AI

Since August 2, 2026, the EU AI Act requires AI-generated content that shows realistic people to be labelled. That is the easy part of transparency. The harder part is what the label leaves open. A label says that an image was generated. It does not say whether a real person is behind it, whether that person agreed, or on what terms. Transparency towards the audience is only half the job. The other half is transparency towards the person whose likeness is used.

On FaceLedger, every licence is documented and every campaign carries a licence proof. Anyone who scans it sees that a verified person stands behind the face, that a licence exists, and for which use and period it was granted. The label says what it is. The licence proof says who agreed.

Self-determination: the face belongs to the person

Self-determination means that a person decides how their own identity is used. In the debate this sounds abstract. In advertising it is a sequence of very specific decisions: which brand, which campaign, which channels, which country, how long, and for how much. A generated face answers none of these questions, because there is nobody to ask.

A licensed face answers all of them. On FaceLedger, every request names the brand, the campaign, the channels, the territory, the term and the amount the person receives. The person can decline without giving a reason. They can exclude up to 25 industries in advance, and they can withdraw their consent, after which no new licences are created and the dataset is deleted. Self-determination is not a statement of intent here. It is a button the person presses, or does not press.

Responsibility: it stays with people

The philosopher Markus Gabriel writes in an essay on the ethics of AI that we must avoid, under all circumstances, delegating our ethics to non-human actors. AI, in his view, is a social technology: it exists only within a human context, and the responsibility for its use stays with the people who build and deploy it.

For advertising this has a plain consequence. The image generator is not responsible for a face that resembles a real person. The company that publishes the campaign is. That is why the question is not whether a tool is ethical, but whether the process around it is. A documented consent, a contract under German law, a verified identity and a proof that can be checked by anyone are what responsibility looks like when it is written down.

What a licence does not solve

Honesty belongs in this debate as well. A licence does not remove bias from training data. It does not give an AI system empathy or judgement. It does not prevent deepfakes made by people who never ask. It solves exactly one part of the problem, and it solves it completely: when a real person appears in AI advertising, that person has agreed, is named in a contract, is paid, and can prove it. Everything else remains the responsibility of the people who use the technology. That is not a weakness of the approach. It is the point.

What this means for your next campaign

If you are planning AI-generated advertising with people in it, the three principles become a checklist. Is the content labelled? Can you show who agreed to the use of the face, for what and for how long? And if someone asks who is responsible, do you have a contract or only a prompt? Companies that can answer all three questions before publication do not need to wait for the ethical debate to be settled. They are already acting on its outcome.